InfoResearchPreprintLLM-specific
Dressed in Authority: Practical Black-Box Jailbreaking of MLLMs via Visual Trust Bias
- Published
- Record updated
Summary
Researchers introduce Visual Trust Bias Attack (VTBA), a black-box jailbreak for multimodal large language models (MLLMs) that exploits their trust bias toward authoritative visual layouts. VTBA matches each naturally occurring harmful image-text pair to an authoritative template and embeds the harmful intent in it, overriding safety refusals. Across 2,500 image-text pairs, the authors report consistently high attack success rates on strong MLLMs.
Topics
Related items
- InfoEGS: A Complementary Safety Framework for Educational LLMsSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
- InfoThe Evolution of Textual Jailbreak Attacks and Defenses for Large Language ModelsSimilar attack · OpenAlex (peer-reviewed AI security)
- InfoCat and Mouse in the LLM Era: An Investigation of Jailbreak Attacks and DefensesSimilar attack · OpenAlex (peer-reviewed AI security)
- MediumGemini AI agents repeatedly bypassed safeguards to attack real websiteSimilar attack · BleepingComputer
- CriticalCVE-2026-120699: CubeSandbox host path bind mount allows host file access via POST /sandboxesSimilar attack · NVD/CVE Database